LIGO used Python (but I don't know everything behind the scenes). Considering the target processors are the same I would suspect the Fortran compilers and optimizers play a part. I did a little hand assembly once because the code the C optimizer did for floating point could not be made to use the full capabilities of the FPU at any settings. Python (numpy) does a pretty fast million point FFT and there is a wrapper for MIT's FFTW.
Here's Nasa's benchmarks. NASA Modeling Guru: Comparing Python, NumPy, Matlab, Fortran, etc.
BTW the MIT developers say somewhere that Intel has "secret" instructions to sell Intel compilers of course (note the winners).
Dunno about secret instructions, but, heck, who else is going to have that kind of characterization of the entire pipeline from memory-on to squeeze all the latencies up just so?
Which might as well be secret instructions.
Edit -- Java was, IMO, the most surprising. And aren't most numpy/scipy modules precompiled C routines? (You can run f2py and c2py, though; I only have second-hand knowledge from close friends, haven't had to get into the weeds myself yet)
Which might as well be secret instructions.
Edit -- Java was, IMO, the most surprising. And aren't most numpy/scipy modules precompiled C routines? (You can run f2py and c2py, though; I only have second-hand knowledge from close friends, haven't had to get into the weeds myself yet)
Last edited:
Yes, I think in general they are. Useless trivia, doing 20*log10() on the output of an FFT does not check for zero (in this case only REALLY zero fails not numerical noise). In 40 years of using simulators I never had a problem but recently found a way to make it break all the time.
It's not the statistics of ABX specifically, although there was a battle some time ago in one of the audio magazines about ABX statistics. One person was complaining that the false negative rate was too high, but that is an artifact of trying to do a low N trial and maintain some modicum of robustness on false positives. He was arguing that we need to relax our standards for what is considered a "real" effect, but given the entire scientific community is moving to higher statistical standards, this is the wrong move. It's better to say it's below our ability to detect than to falsely claim s positive conclusion on dodgy data, even if that goes against the grain of our intuition and the way the lay community communicates.
The moral of the story is not that ABX is inherently bad, but trying to elucidate anything from a small dataset is a recipe for failure.
In terms of an experimental design, ABX may have a higher cognitive load than other protocols, and thus be less sensitive. But that isn't definitive. High N trials require a lot out of the tested individuals (victims), and really have to be spaced out to avoid fatigue.
The moral of the story is not that ABX is inherently bad, but trying to elucidate anything from a small dataset is a recipe for failure.
In terms of an experimental design, ABX may have a higher cognitive load than other protocols, and thus be less sensitive. But that isn't definitive. High N trials require a lot out of the tested individuals (victims), and really have to be spaced out to avoid fatigue.
Last edited:
Then you are no statistician. No reputable paper in any field dealing with humans would ever be accepted if it did not us double blind testing, i.e. ABX. What you are claiming is simply not true.
Correct! The false negatives claims that subjectivists hang their hat on is totally blown out of proportion. In enough tests it is not a factor.
Finding the "limit" is not of much interest to me as this requires specialized signals and techniques. What I want to know is when nonlinearity influences the music that we hear and that is an entirely different aspect.
Unless this effect were pronounced and occurred with many people I would discount it as fluke event and not worthy of interpretation.
This depends entirety on what you mean by "low level" and what orders of nonlinearity we are talking about. As I said, I can create a nonlinearity at .1 % THD that everyone will hear and another at 20% THD that no one will hear. What does that mean?
No one is going to research this!! No one paid attention when we did our preliminary study. The response was simply "We all know that THD doesn't work, but its so easy to do!!!"
PS. FORTRAN is still my main programming language, but only for crunching numbers. For GUI I use VB.Net but call the huge world of mathematical subroutines that FORTRAN offers. Its complex number abilities are unparalleled (or I should say "paralleled", since FORTRAN is now a totally parallel language.)
Agreed. The problem I would like to address is that many people will insist than no human can hear less than .1% THD, that there is no point designing amplifiers much cleaner than that, and anyone who claims to hear lower levels of distortion in music reproduction systems is a liar or a self-deluded hallucinating fool. The same people like to insist the *only* possible proof that someone can hear less than .1% THD is to present ABX test results using Foobar2000 (because that's the only freeware ABX test available).
The problem with providing the "proof" is that many people have great difficulty not getting distracted by the ABX process to the extent that they significantly fail to hear as consistently during testing as they do normal listening, but only for distortion levels below .1% THD. I found that a sorting test seems to work for me without interfering with how I normally hear distortion in music. I also believe that it would probably be possible to produce a sorting test that is just hard to game or cheat as current implementations of ABX. We just don't have a free one, and no one has published research showing it is a viable alternative to ABX.
Futher, I think if we had a test that interfered less with sensitive hearing than the ABX tests we have now, we might find more people may hear and find objectionable low levels non-linear distortion when listening to recorded music than some people currently believe.
Maybe I over spoke about self-delusional fool thing. It hasn't been that bad around here lately. Probably more accurate to say some people have been accused of imagining things that could possibly be real. And the matter is complicated by a few claims that seem physically absurd. Hard to sort out what may be real from what may be imagined. I would vote for more research.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.